Engineering Circular RNA for Potent and Stable Translation in Eukaryotic Cells

ARTICLE

DOI: 10.1038/s41467-018-05096-6 OPEN Engineering circular RNA for potent and stable translation in eukaryotic cells

R. Alexander Wesselhoeft 1,2, Piotr S. Kowalski3 & Daniel G. Anderson1,3,4,5

Messenger RNA (mRNA) has broad potential for application in biological systems. However, one fundamental limitation to its use is its relatively short half-life in biological systems. Here we develop exogenous circular RNA (circRNA) to extend the duration of protein expression fi 1234567890():,; from full-length RNA messages. First, we engineer a self-splicing intron to ef ciently circularize a wide range of RNAs up to 5 kb in length in vitro by rationally designing ubiquitous accessory sequences that aid in splicing. We maximize translation of functional protein from these circRNAs in eukaryotic cells, and we find that engineered circRNA purified by high performance liquid chromatography displays exceptional protein production qualities in terms of both quantity of protein produced and stability of production. This study pioneers the use of exogenous circRNA for robust and stable protein expression in eukaryotic cells and demonstrates that circRNA is a promising alternative to linear mRNA.

1 David H. Koch Institute for Integrative Cancer Research, Massachusetts Institute of Technology, Cambridge, MA 02139, USA. 2 Department of Biology, Massachusetts Institute of Technology, Cambridge, MA 02139, USA. 3 Department of Chemical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA. 4 Institute for Medical Engineering and Science, Massachusetts Institute of Technology, Cambridge, MA 02139, USA. 5 Harvard and MIT Division of Health Science and Technology, Massachusetts Institute of Technology, Cambridge, MA 02139, USA. Correspondence and requests for materials should be addressed to D.G.A. (email: [email protected])

NATURE COMMUNICATIONS | (2018) 9:2629 | DOI: 10.1038/s41467-018-05096-6 | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05096-6

ircular RNAs (circRNAs) endogenous to eukaryotic cells the 5′ and 3′ ends of the precursor RNA with the aim of bringing have drawn increasing interest due to their prevalence and the 5′ and 3′ splice sites into proximity of one another (Fig. 1b, C 1 range of potential biological functions . Most circRNAs Supplementary Data 1). Addition of these homology arms – found in nature are generated through backsplicing2 4 and appear increased splicing efficiency from 0 to 16% for weak homology to fulfill noncoding roles1, 5, 6. However, it has been demonstrated arms and to 48% for strong homology arms as assessed by dis- that some circRNAs endogenous to Drosophila and humans appearance of the precursor RNA band (Fig. 1c). To ensure that encode proteins7, 8. In addition to having protein-coding poten- the major splicing product was circular, we treated the splicing tial, endogenous circRNAs lack the free ends necessary for reaction with RNase R (Fig. 1d, Supplementary Fig. 1a). exonuclease-mediated degradation, rendering them resistant to Sequencing across the putative splice junction of RNase R-treated several mechanisms of RNA turnover and granting them exten- splicing reactions revealed ligated exons, and digestion of the ded lifespans as compared to their linear mRNA counterparts2, 9. RNase R-treated splicing reaction with oligonucleotide-targeted For this reason, circularization may allow for the stabilization of RNase H produced a single band in contrast to two bands yielded mRNAs that generally suffer from short half lives10, 11 and may by RNase H-digested linear precursor (Fig. 1d, e, Supplementary therefore improve the overall efficacy of exogenous mRNA in a Fig. 1a). These data show that circRNA is a major product of variety of applications12. these splicing reactions and that agarose gel electrophoresis allows Previous efforts to increase mRNA stability include the use of for simple and effective separation of circular splicing products untranslated regions such as those present in the native beta from linear precursor molecules, nicked circles, splicing inter- globin mRNA, methylguanosine cap analogs to protect mRNA mediates, and excised introns. from decapping enzymes, nucleoside modification, and codon optimization. These strategies have yielded modest improvements fi 10, 13–16 Spacer sequences improve circularization ef ciency. In order to in RNA stability , and thus alternative approaches to RNA further improve the efficiency of circRNA generation from the stabilization, such as circularization, are desirable. However, the fi self-splicing precursor RNA, we considered other factors that may ef cient circularization of long in vitro transcribed (IVT) RNA, influence successful circularization. The 3′ PIE splice site is the purification of circRNA, and the adequate expression of fi proximal to the IRES, and because both sequences are highly protein from circRNA are signi cant obstacles that must be structured, we hypothesized that sequences within the IRES may overcome before their protein-coding potential can be realized. In interfere with the folding of the splicing ribozyme, either proxi- this study, we present an engineering approach to generating mally at the 3′ splice site or distally at the 5′ splice site through exogenous circRNAs for potent and durable protein expression in long-distance contacts. In order to allow these structures to fold eukaryotic cells. independently, we designed a series of spacers between the 3′ PIE splice site and the IRES that we predicted would either permit or disrupt splicing (Fig. 2a, Supplementary Data 1). Permissive Results spacers were designed to conserve secondary structures present Long RNA circularization is assisted by homology. There are within intron sequences that may be important for ribozyme three general strategies for exogenous RNA circularization: che- activity, while the disruptive spacer was designed to disrupt mical methods using cyanogen bromide or a similar condensing sequences in both intron halves, especially the 5′ half. The agent, enzymatic methods using RNA or DNA ligases, and 17–19 addition of spacer sequences predicted to permit splicing ribozymatic methods using self-splicing introns . A ribozy- increased splicing efficiency from 46 to 87% (P1 and P2), while matic method utilizing a permuted group I catalytic intron has the addition of a disruptive spacer sequence completely abrogated been reported to be more applicable to long RNA circularization + 17 splicing (Fig. 2b). This improved construct, containing both and requires only the addition of GTP and Mg2 as cofactors . homology arms and rationally designed spacers, was able to cir- This permuted intron-exon (PIE) splicing strategy consists of fl 20 cularize RNA approaching 5 kb in length (Supplementary fused partial exons anked by half-intron sequences . In vitro, Fig. 1b). these constructs undergo the double transesterification reactions characteristic of group I catalytic introns, but because the exons are already fused they are excised as covalently 5′ to 3′ linked Anabaena catalytic intron is superior for circularization.We circles17 (Fig. 1a). Using this strategy as a starting point for also explored the use of an alternative group I catalytic intron from the Anabaena pre-tRNA20. Here we applied the same creating a protein coding circular RNA, we inserted a 1.1 kb fi sequence containing a full-length encephalomyocarditis virus optimization techniques that we used to increase the ef ciency of (EMCV) IRES, a Gaussia luciferase (GLuc) message, and two the permuted T4 phage intron splicing reaction. Interestingly, short regions corresponding to exon fragments (E1 and E2) of the during our optimizations we noted that switching from the T4 PIE construct between the 3′ and 5′ introns of the permuted catalytic intron to the Anabaena catalytic intron may have resulted in the weakening of a short stretch of internal homology group I catalytic intron in the thymidylate synthase (Td) gene of ′ the T4 phage21 (Fig. 1a, Supplementary Data 1). Precursor RNA between the IRES and the 3 end of the coding region, which may was synthesized by run-off transcription and then heated in the have aided in the formation of an isolated splicing bubble (Fig. 2c, Fig. 3a). Strengthening this internal homology further increased presence of magnesium ions and GTP to promote circularization, fi essentially as described previously for the circularization of splicing ef ciency from 84 to 95% using the permuted Anabaena shorter RNAs21. However, we were unable to obtain splicing catalytic intron (Fig. 2c, d, Supplementary Data 1). We also found products. We speculated that long intervening regions between that use of the Anabaena catalytic intron resulted in a 37% reduction in circRNA nicking compared to the T4 catalytic intron splice sites which may reduce the ability of the splice sites to fi interact with one another and form a stable complex, thus (Fig. 2d, e). Due to increased splicing ef ciency and intact cir- reducing splicing efficiency. Indeed, the intervening region cRNA output, the engineered Anabaena PIE system proved to be between the 5′ and 3′ splice sites of native group I introns is on overall superior to the engineered T4 PIE system (Fig. 2e, Sup- average 300–500 nucleotides long22, while the intervening region plementary Fig. 2a). of the engineered RNA that we constructed was two to four-fold longer. Therefore we designed perfectly complementary ‘homol- Autocatalytic sequences are compatible with coding regions. ogy arms’ 9 (weak) or 19 (strong) nucleotides in length placed at Internal homology between exon 2 and the GLuc coding sequence

2 NATURE COMMUNICATIONS | (2018) 9:2629 | DOI: 10.1038/s41467-018-05096-6 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05096-6 ARTICLE

E1 E2 5′ Intron 3′ Intron

Insert

0 3.6 E2 E1 3′ Intron 5′ Intron a) No homology arms

E2 EMCV IRES Gaussia luciferase E1 3′ Intron 5′ Intron

G-OH

Precursor E2 EMCV IRES Gaussia luciferase E1 b) Weak homology arms c) Strong homology arms 3′ Intron 5′ Intron cd(–) Weak Strong C C+R C+R+H U U+H

G- Circular Circular Precursor OH Precursor Nicked Intermediates Nicked

Bound introns Bound introns

Products e G-

′ ′ 3 Intron 5 Intron 200 210 220 230 Splice junction 5′ Exon 3′ Exon

Fig. 1 Permuted intron-exon splicing and addition of homology arms. a Schematic diagram showing permuted intron-exon construct design and mechanism of splicing. The group I catalytic intron of the T4 phage Td gene is bisected in such a way to preserve structural elements critical for ribozyme folding. Exon fragment 2 is then ligated upstream of exon fragment 1, and a coding region roughly 1.1 kb in length is inserted between the exon-exon junction. During splicing, the 3′ hydroxyl group of a guanosine nucleotide engages in a transesterification reaction at the 5′ splice site. The 5′ intron half is excised, and the freed hydroxyl group at the end of the intermediate engages in a second transesterification at the 3′ splice site, resulting in circularization of the intervening region and excision of the 3′ intron. b RNAFold predictions of precursor RNA secondary structure for homology arm design. Colors denote base pairing probability, with red indicating higher probability. Without homology arms, no base pairing is predicted to occur between the ends of the precursor molecule. c Agarose gel demonstrating the effect of homology arms on splicing. Putative circRNA runs at a higher molecular weight than heavier precursor RNA, as indicated. (−): no homology arms. Weak: weak homology arms, 9 nt. Strong: strong homology arms, 19 nt. d Agarose gel confirmation of precursor RNA circularization. C: precursor RNA (with strong homology arms) subjected to circularization conditions. C + R: Lane C, digested with RNase R. C + R + H: Lane C + R, digested with oligonucleotide-guided RNase H. U: precursor RNA not subjected to circularization conditions. U + H: Lane U, digested with oligonucleotide-guided RNase H. e Sanger sequencing output of RT-PCR across the splice junction of the sample depicted in lane C + R from (d)

rendered the optimized Anabaena PIE system incompatible with included homology arms at the 5′ and 3′ ends of the precursor non-GLuc intervening regions. To adapt our circRNA construct molecule. Between these sequences we inserted an EMCV IRES, for efficient circularization of a variety of long intervening RNA as well as coding regions for five different proteins, including sequences, we de novo designed a pair of spacer sequences based Gaussia luciferase (total length: 1289nt), Firefly luciferase on our understanding of the parameters that affect permuted (2384nt), eGFP (1451nt), human erythropoietin (1313nt), and catalytic group I intron splicing efficacy. These spacer sequences Cas9 endonuclease (4934nt). We were able to achieve circular- were engineered with three priorities: (1) to be unstructured and ization of all five RNA sequences (Fig. 3b, Supplementary Data 1); non-homologous to the proximal intron and IRES sequences; (2) circularization efficiency matched that of our stepwise-designed to separate intron and IRES secondary structures so that each construct (Fig. 2e) and was highly reproducible between inserts element is able to fold and function independently of one but was also dependent on size, with long RNAs less efficiently another; and (3) to contain a region of spacer–spacer com- circularized (Supplementary Fig. 3a). In addition we found that plementarity to promote the formation of a sheltered splicing long circRNAs were more prone to nicking in the presence of bubble containing catalytic intronic sequences flanked by two magnesium ions, resulting in accumulation of nicked circRNA regions of homology (Fig. 3a, Supplementary Data 1). We also during and after in vitro transcription and RNase R digestion

NATURE COMMUNICATIONS | (2018) 9:2629 | DOI: 10.1038/s41467-018-05096-6 | www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05096-6

a b (–) D P1 P2

Circular Precursor Nicked

Bound introns a) No spacer b) Disruptive spacer c) Permissive spacer 1 d) Permissive spacer 2 Free introns

c deAna1.0 Ana2.0T4 Ana2.0

Circular Circular Precursor Precursor Nicked Nicked

03.6 Bound introns Bound introns

Free introns Free introns Anabaena 1.0 Anabaena 2.0

Fig. 2 Spacer design and splicing using the Anabaena autocatalytic intron. a RNAFold predictions of precursor RNA secondary structure in the context of designed spacers. Secondary structures potentially important for ribozyme function are identified by black arrows. b Agarose gel demonstrating the effect of spacers on splicing. (−): no spacer. D: disruptive spacer. P1: permissive spacer 1. P2: permissive spacer 2. c RNAFold predictions of precursor RNA secondary structure for internal homology region design. Lack of significant internal homology (Anabaena 1.0) and introduced internal homology (Anabaena 2.0) indicated by black arrows. Splicing bubble indicated as the region between homology arms and internal homology regions that contains the splicing ribozyme. d Agarose gel demonstrating the effect of internal homology on splicing. e Agarose gel comparing the optimized T4 phage splicing reaction to the optimized Anabaena splicing reaction. Anabaena intron halves are of roughly equal lengths, and are less likely to remain associated after splicing in comparison to the T4 phage intron halves despite stronger homology arms which reduced the overall yields and the purity of the RNase R- linear mRNA technology it is desirable to maximize protein treated sample (Fig. 3b, c, Supplementary Fig. 3a). RNase R did expression. Cap-independent translation mediated by an IRES not fully digest resistant Anabaena introns (Fig. 3b, bottom can exhibit varying levels of efficiency depending on cell context bands) or rare circular concatenations representing approxi- and is generally considered less efficient than cap-dependent mately 1.07% of splicing reactions (Fig. 3b, faint top bands). translation when included in bicistronic linear mRNA23. Simi- larly, the polyA tail stabilizes and improves translation initiation fi Exogenous circRNA is efficiently translated. It has been ef ciency in linear mRNA through the actions of polyadenylate 24, 25 fi demonstrated that endogenous circRNA may produce small binding proteins . However, the ef ciency of different IRES quantities of protein7. As a means of assessing the ability of sequences and the inclusion of a polyA tract within the context of engineered circRNAs to produce protein, we transfected RNase circRNA has not been investigated. We replaced the EMCV IRES ′ R-digested splicing reactions of each construct into human with 5 UTR sequences from several viral transcripts that contain embryonic kidney cells (HEK293). Transfection of Gaussia or known or putative IRESs, as well as several other putative IRES 26 Firefly luciferase circRNA resulted in robust production of sequences (Supplementary Data 1, Supplementary Fig. 4a) .We functional protein as measured by luminescence (Fig. 3d, f). found that the IRES from Coxsackievirus B3 (CVB3) was 1.5-fold Likewise, we were able to detect human erythropoietin in cell more effective than the commonly adopted EMCV IRES in culture media from transfection of erythropoietin circRNA, and HEK293 cells (Fig. 4a). Because secondary structures proximal to observe EGFP fluorescence from transfection of EGFP circRNA the IRES, including within the coding region that directly follows (Fig. 3e, g). Co-transfection of Cas9 circRNA with sgRNA the IRES, have the potential to disrupt IRES folding and trans- directed against GFP into HEK293 cells constitutively expressing lation initiation, we also tested selected viral IRES sequences in fl GFP resulted in ablated fluorescence in up to 97% of cells in the context of Fire y luciferase. While the CVB3 IRES was still fi comparison to an sgRNA-only control (Fig. 3h, Supplementary superior to all others, the ef cacy of several other IRESs, most Fig. 3d, e). Because RNase R digestion of splicing reactions is not notably the Poliovirus IRES, was dramatically altered (Fig. 4a). always complete and precursor RNA contains a functional IRES, We then tested whether the addition of an internal polyA we also created a splice site deletion mutant of the GLuc construct sequence or a polyAC spacer control to IRES sequences that to measure the potential contribution of impurities to protein showed the ability to drive protein production above background expression. When transfected at equal weight quantities to levels from engineered circRNA would alter protein expression. RNase-R digested splicing reactions, this splice site deletion We found that both sequences improved expression in all con- mutant produced a barely detectable level of protein (Supple- structs, possibly due to greater unstructured separation between – mentary Fig. 3b, c). the beginning of the IRES sequence and the exon exon splice junction, which is predicted to maintain a stable structure (Fig. 4b, Supplementary Fig. 4b). This greater degree of The CVB3 IRES is superior in the context of circRNA.To unstructured separation may reduce steric hindrance occluding establish exogenous circRNA as a reliable alternative to existing

4 NATURE COMMUNICATIONS | (2018) 9:2629 | DOI: 10.1038/s41467-018-05096-6 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05096-6 ARTICLE

a b GLuc hEpo EGFP FLuc Cas9 R– R+ R– R+ R– R+ R– R+ R– R+

5′ Internal homology 3′ Internal homology Splicing bubble Fold

E2EMCV IRES Coding region E1 3′ Homology arm 5′ Homology arm 5′ Spacer 3′ Spacer

cdef Circular RNA Yields 6 5 6 4.0 × 10 * 2.0 × 10 1.5 × 10 1.0 * * 3.0 × 106 1.5 × 105 1.0 × 106 2.0 × 106 1.0 × 105 0.5 5.0 × 105 6 4 Yield (%) Yield 1.0 × 10 5.0 × 10 activity (RLU) activity (RLU) Firefly luciferase

Gaussia luciferase ND 0.0 0 0 0 hEpo GLuc Erythropoietin (mIU/mL) GLuc hEpo hEpo FLuc GLuc FLuc+R GLuc+RhEpo+REGFP+R Cas9+R

g h circCas9– circCas9+ hEpo EGFP 500 97.5% 97.0%

400 300

300 400 um 400 um 200 Events Events 200 2.5% 3.0% 100 100

0 0 –103 0 103 104 105 –103 0 103 104 105 GFP GFP

Fig. 3 Evaluation of circularization efficacy and translation for a range of protein-coding circRNAs generated from de-novo engineered precursor RNA. a Schematic diagram showing elements of the engineered self-splicing precursor RNA design. b Agarose gel of precursor RNA containing an EMCV IRES and a variable insert including Gaussia luciferase (GLuc), human erythropoietin (hEpo), EGFP, Firefly luciferase (FLuc), or Cas9 coding regions after circularization and recircularization (R−). CircRNA was enriched by RNase R degradation (R+). c Approximate circRNA yields from treatment of 20 µgof splicing reaction with RNase R, as assessed by spectrophotometry (n = 3). d Luminescence in the supernatant of HEK293 cells 24 h after transfection with circRNA coding for GLuc (n = 4). e Expression of human erythropoietin in the supernatant of HEK293 cells 24 h after transfection with circRNA coding for hEpo as assessed by solid-phase sandwich ELISA (n = 4). f Luminescence in the lysate of HEK293 cells 24 h after transfection with circRNA coding for FLuc (n = 4). g GFP fluorescence in HEK293 cells 24 h after transfection with circRNA coding for EGFP (scale bar: 400 µm). h FACS analysis demonstrating GFP ablation in HEK293-EF1a-GFP cells 4 days after transfection with sgGFP alone (circCas9−) or co-transfected with circRNA coding for Cas9 (circCas9+), indicated by the appearance of a GFP-negative cell population (all data presented as mean + SD, *p < 0.05 (Welch’s t-test)) initiation factor binding to IRES structures. In the case of EMCV innate cellular immune responses. It has been shown that removal and Poliovirus IRESs, polyA sequences improved expression of dsRNA by HPLC eliminates immune activation and improves beyond the improvement seen with an unstructured polyAC translation of linear nucleoside-modified IVT mRNA16. However, spacer. This may suggest that the association of polyadenylate no scalable methods have been reported for purification of cir- binding proteins could enhance IRES efficiency. After selecting cRNA from byproducts of IVT and circularization reactions, the most effective polyA or polyAC construct for each IRES, we which include dsRNA and triphosphate-RNA that may engage explored IRES efficacy in different cell types, including human RNA sensors and induce a cellular immune response16. While the cervical adenocarcinoma (HeLa), human lung carcinoma (A549), complete avoidance of nicked circRNA was untenable due to mild and immortalized mouse pancreatic beta cells (Min6). We found degradation during processing, we were able to obtain sub- that IRES efficacy varied depending on cell type, but the CVB3 stantially pure (90% circular, 10% nicked) circRNA using gel IRES was superior in all types tested (Fig. 4c, d). Deletion of extraction for small quantities and size exclusion HPLC for larger relatively short proximal or distal CVB3 IRES sequences resulted quantities of splicing reaction starting material (Fig. 5a, b). In in dramatic loss of protein expression (Supplementary Fig. 4c). both cases, we followed purification with RNase R treatment to eliminate the majority of degraded RNA. When comparing the protein expression of gel extracted or HPLC purified circRNA to CircRNA can be purified from splicing reactions using HPLC. RNase-R digested splicing reactions, we found that HPLC pur- Purity of circRNA preparations is another factor essential for ification was a superior method of purification that surpassed maximizing protein production from circRNA and for avoiding RNAse R digestion alone (Fig. 5b, c).

NATURE COMMUNICATIONS | (2018) 9:2629 | DOI: 10.1038/s41467-018-05096-6 | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05096-6

ab* 4.0 × 106 *

4.0 × 106 4.0 × 106 Firefly luciferase activity (RLU) 3.0 × 106

6 6 3.0 × 10 3.0 × 10 *

6 * 2.0 × 10 2.0 × 106 2.0 × 106 * * * * 1.0 × 106 * * 1.0 × 106 1.0 × 106 * *

Gaussia luciferase activity (RLU) Gaussia luciferase * * 0 0 0

(–) pA (–) pA (–) pA (–) pA (–) pA (–) pA PV pAC pAC pAC pAC pAC pAC HAV CVB3 EV71 CSFV HRV2 ABPV HTLV Control Control EMCV CVB3 EV71 EMCV PV CSFV HRV2

c 6 d 5.0 × 106 3.0 × 10 2.0 × 106 5.0 × 105 HeLa GLuc activity (RLU) Min6 GLuc activity (RLU) 5 4.0 × 106 4.0 × 10 1.5 × 106 2.0 × 106 3.0 × 106 3.0 × 105 1.0 × 106 2.0 × 106 2.0 × 105 1.0 × 106 5.0 × 105 1.0 × 106 1.0 × 105 A549 GLuc activity (RLU) HEK293 GLuc activity (RLU) activity (RLU) Gaussia luciferase 0 0 0 0

PV pA PV pA Control Control EV71 pA EMCV pA EV71 pA CVB3 pAC CSFV pAC HRV2 pAC CVB3 pAC EMCV pA CSFV pAC HRV2 pAC

Fig. 4 IRES efﬁcacy in varying cell and sequence contexts. a Luminescence in the supernatant of HEK293 cells 24 h after transfection with circRNA containing a panel of viral 5′ UTR IRES sequences in GLuc (left axis, dark purple with black outline) and FLuc (right axis, light purple with gray outline) contexts. b Luminescence in the supernatant of HEK293 cells 24 h after transfection with circRNA containing an added polyA(30) or polyAC(30) spacer sequence separating the IRES from the splice junction. (−): no spacer. pAC: 30nt spacer consisting of adenosines and cytosines. pA: 30nt spacer consisting of adenosines. c Luminescence in the supernatant of HEK293 (left axis, dark purple with black outline) and HeLa (right axis, light purple with gray outline) cells 24 h after transfection with the most effective circRNAs by IRES in (b). d Luminescence in the supernatant of A549 (left axis, dark purple with black outline) and Min6 (right axis, light purple with gray outline) cells 24 h after transfection with the most effective circRNAs by IRES in (b) (all data presented as mean + SD, n = 4, *p < 0.05 (Welch’s t-test))

Protein production is comparable to linear mRNA.Itis lifetime, while weaker initial expression did not abrogate the unknown whether exogenous circRNA translation efficiency is observed stability phenotype (Fig. 6c, Supplementary Fig. 6c). In comparable to that of linear mRNA, and whether circRNA pro- HeLa cells, circRNA exhibited a protein production half-life of tein production exhibits differences in stability. Using HPLC- 116 h, while the half-lives of protein production from the purified engineered circRNA, we compared the stability and unmodified and modified linear mRNA were approximately 44 efficacy of Gaussia luciferase-coding circRNA (CVB3-GLuc-pAC) and 49 h, respectively (Fig. 6d). This again resulted in sub- to equimolar quantities of a canonical unmodified 5′ stantially more protein production from circRNA over its lifetime methylguanosine-capped and 3′ polyA-tailed linear GLuc mRNA, compared to both the unmodified and modified linear mRNAs as well as a commercially available nucleoside-modified (pseu- (Fig. 6c). We did not observe this enhanced expression or stability douridine, 5-methylcytosine) linear GLuc mRNA (Trilink). Pro- from a capped and polyadenylated circRNA precursor containing tein production assessed by luminescence 24 h post-transfection all accessory sequences (Supplementary Fig. 6a, b). revealed that circRNA produced 811.2% more protein than the unmodified linear mRNA at this early time point in HEK293 cells (Fig. 6a). Interestingly, circRNA also produced 54.5% more Discussion protein than the modified mRNA, demonstrating that nucleoside Obtaining stable protein production from exogenous mRNA has modifications are not necessary for robust protein production been a long-standing goal of mRNA biotechnology. Previous from circRNA. Similar results were obtained in HeLa cells efforts to increase mRNA stability have yielded modest – (Fig. 6a) and using optimized circRNA coding for human ery- improvements10, 13 16. The possibility of adapting circular RNA thropoietin in comparison to linear mRNA modified with 5- for the purpose of stabilizing mRNA has been stifled by low methoxyuridine (Supplementary Fig. 5a, b). Luminescence data circRNA production efficiency, difficulty of purification, and collected over 6 days showed that protein production from cir- weak protein expression. Indeed, these obstacles must be over- cRNA was extended relative to that from the linear mRNA in come before the stability of protein production from circRNA can HEK293 cells, with circRNA exhibiting a protein production half- be fully assessed. Previously, the permuted group 1 catalytic life of 80 h, while the half-lives of protein production from the intron-based system has been used to circularize a wide range of unmodified and modified linear mRNAs were approximately 43 short RNA sequences in vitro, with circularization efficiencies and 45 h, respectively (Fig. 6b). Due to increased expression or reported to reach 90% for RNAs between 58 and 124 nt20, 21. stability, circRNA also produced substantially more protein than Longer RNAs (up to 1.5 kb) have recently been circularized using both the unmodified and modified linear mRNAs over its this method, but circRNA yields from these constructs have not

6 NATURE COMMUNICATIONS | (2018) 9:2629 | DOI: 10.1038/s41467-018-05096-6 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05096-6 ARTICLE

a b mAU 1600 C +R +G,R +H,R 1400 1200 Concatenations 1000 800 600 Circular 400 Nicked 200 0 0 Retention time 5 10 15 min

mAU

1750 1500 1250 1000 750 Free introns 500 250 0 0 Retention time 5 10 15 min

c 3.0×106 Concatenations * * * Circular Precursor 6 Nicked 2.0×10

1.0×106

Bound introns Gaussia luciferase activity (RLU) Gaussia luciferase

0 Free introns C C+R Control C+G/R C+H/R

Fig. 5 HPLC purification of circRNA from splicing reactions. a HPLC chromatogram of linear GLuc RNA (top) and a CVB3-GLuc-pAC splicing reaction (middle). Agarose gel of collected fractions (bottom). b Agarose gel of CVB3-GLuc-pAC purified by different methods. C: splicing reaction. +R: splicing reaction treated with RNase R. +G,R: splicing reaction gel extracted, and then treated with RNase R. +H,R: splicing reaction HPLC purified, and then treated with RNase R. c Luminescence in the supernatant of HEK293 cells 24 h after transfection with the CVB3-GLuc-pAC splicing reactions purified by different methods as noted in (b) (data presented as mean + SD, n = 4, *p < 0.05 (Welch’s t-test)) been reported28. The engineered permuted group 1 catalytic autohydrolysis27, significantly reduces yields and is another intron-based system described herein permits the circularization deficiency that requires improvement. Longer circRNAs have of sequences up to 5 kb in length, significantly longer than pre- more opportunities to auto-hydrolyze simply based on their size. viously reported. Moreover, we are able to obtain nearly 100% This degradation is more evident for circRNA than it is for linear circularization efficiency for a range of inserted coding regions RNA because single nicks in circRNA manifest as a single band with diverse sequence compositions. We show that optimized on a gel, whereas single nicks in linear RNA manifest as a smear circRNA is capable of producing large quantities of protein and that is diluted over a range of molecular weights. This phenom- also that it can be effectively be purified by HPLC. Finally, we enon is visible in Fig. 3b, lane 7 (FLuc, R-, as well as several other demonstrate that circRNA can produce greater quantities of lanes); two major bands exist in this lane, with the top major protein for a longer duration than the unmodified and modified band representing intact circRNA while the bottom major band linear RNAs used in this study, providing evidence that circRNA represents nicked circRNA. No smear extends from the intact holds potential as an alternative to mRNA for the stable circRNA band, while a smear extends from the nicked circRNA expression of protein. However, more work needs to be done to band. The nicked circRNA band represents single nicks that fully investigate the potential of circRNA for therapeutic and occur at random positions in an intact circRNA, while the smear non-therapeutic applications, including further optimizations of that extends from it represents additional nicks beyond the first protein translation from circRNA and a comprehensive study of that occur at random positions in the linearized nicked circRNA the translation efficiency and stability of circRNA in other cell and thus give rise to degradation products of varying molecular types and tissues. The nicking of longer circRNAs that we weights. We do not expect circRNA nicking to present challenges observed during the production of circRNA (Supplementary beyond those that already exist for maintaining the stability of Fig. 2a), likely mediated by magnesium-catalyzed linear RNA in solution and during processing.

NATURE COMMUNICATIONS | (2018) 9:2629 | DOI: 10.1038/s41467-018-05096-6 | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05096-6

ab 6 * HEK293 6.0 × 10 * * 6.0 × 106 *

activity (RLU) Circular 6 6 HeLa GLuc 1.0 4.0 × 10 4.0 × 10 Mod. linear Unmod. linear

2.0 × 106 2.0 × 106 activity (RLU) HEK293 GLuc 0.5

0 0 activity (RLU) Normalized GLuc

Control Circular Control Circular Mod. linear Mod. linear 0.0 Unmod. linear Unmod. linear 123456 Days

c d HeLa

Circular 7 * 7 2.0 × 10 * 2.5 × 10 1.0 Mod. linear * Cumulative HeLa GLuc 7 Unmod. linear 7 2.0 × 10 1.5 × 10 * activity (RLU) 1.5 × 107 1.0 × 107 1.0 × 107 0.5

6 activity (RLU) activity (RLU) 5.0 × 10 6 5.0 × 10 Normalized GLuc

Cumulative HEK293 GLuc Cumulative 0 0 0.0 123456 Control Circular Control Circular Mod. linear Mod. linear Days Unmod. linear Unmod. linear

Fig. 6 Translation efficacy of circRNA compared to linear mRNA. a Luminescence in the supernatant of HEK293 (left, black outline) and HeLa (right, gray outline) cells 24 h after transfection with CVB3-GLuc-pAC circRNA or modified or unmodified linear GLuc mRNA (n = 4 HEK293, n = 3 HeLa). b Luminescence in the supernatant of HEK293 cells starting 24 h after transfection with CVB3-GLuc-pAC circRNA or modified or unmodified linear GLuc mRNA and continuing for 6 days (n = 4). c Relative cumulative luminescence produced over 6 days by HEK293 (left, black outline) and HeLa (right, gray outline) cells transfected with CVB3-GLuc-pAC circRNA or modified or unmodified linear GLuc mRNA (n = 4 HEK293, n = 3 HeLa). d Luminescence in the supernatant of HeLa cells starting 24 h after transfection with CVB3-GLuc-pAC circRNA or modified or unmodified linear GLuc mRNA and continuing for 6 days (n = 3). (all data presented as mean + SD, *p < 0.05 (Welch’s t-test))

It has recently been demonstrated that transfection of cells with thorough investigation of suitable vehicles for the delivery of exogenous circRNA results in the activation of antiviral gene circRNA in vitro and in vivo is required. products such as OAS, PKR, and RIG-I, and that RIG-I is likely to be the principal component responsible for the cellular response Methods against circRNA28. We observed our circRNA preparations to Cloning and mutagenesis. Protein coding, group I self-splicing intron, and IRES elicit a significantly stronger IFN-ɑ and IL-6 cytokine response in sequences were chemically synthesized (Integrated DNA Technologies) and cloned A549 cells than untransfected controls, although the response was into a PCR-linearized plasmid vector containing a T7 RNA polymerase promoter fi by Gibson assembly using a NEBuilder HiFi DNA Assembly kit (New England comparable to that from unmodi ed linear mRNA (Supple- Biolabs). Spacer regions, homology arms, and other minor alterations were mentary Fig. 7a). We also found a 30-fold induction of IFN-β1 introduced using a Q5 Site Directed Mutagenesis Kit (New England Biolabs). and a 3-fold induction of RIG-I transcripts upon transfection of Primers for this and the following methods can be found in Supplementary Data 1. HeLa cells with circRNA, compared to mock transfection (Sup- plementary Fig. 7b). The 3-fold induction of RIG-I mRNA we circRNA design and purification. RNA structure was predicted using RNAFold18. observed was significantly lower than the 500-fold induction Modified linear GLuc mRNA was obtained from Trilink Biotechnologies and 28 fi consisted of a codon optimized GLuc coding region, a proprietary synthetic 5′ reported by Chen et al. in HeLa cells and was not suf cient to untranslated region, an alpha globin 3′ untranslated region, a cap 1 structure, a suppress protein translation and the stability of protein expres- 120-nucleotide polyA tail, and complete replacement of uridine and cytosine along sion from circRNA in this cellular context (Fig. 6d). The observed the entire mRNA with pseudouridine and 5-methylcytosine, respectively. Modified differences in RIG-I activation could be attributed to differences hEpo mRNA was also obtained from Trilink Biotechnologies and was structurally fi identical to the Trilink GLuc mRNA described above, except that it was modified in construct design and circRNA puri cation methods. To further with 5-methoxyuridine and the coding region coded for human erythropoietin. suppress an immune response directed against circRNA, methods Unmodified linear RNA consisted of a GLuc or hEpo coding region but did not such as nucleoside modification could potentially be incorporated include specific untranslated regions. Unmodified linear mRNA or circRNA pre- into circRNA design13, 16. Additional analysis of exogenous cir- cursors were synthesized by in-vitro transcription from a linearized plasmid DNA template using a T7 High Yield RNA Synthesis Kit (New England Biolabs). After cRNA immunogenicity in the context of potential applications is in vitro transcription, reactions were treated with DNase I (New England Biolabs) required. for 20 min. After DNase treatment, unmodified linear mRNA was column purified Finally, we show that circRNA is efficiently delivered to cells using a MEGAclear Transcription Clean-up kit (Ambion). RNA was then heated to in vitro by a cationic lipid transfection reagent. However, a 70 °C for 5 min and immediately placed on ice for 3 min, after which the RNA was capped using mRNA cap-2′-O-methyltransferase (NEB) and Vaccinia capping

8 NATURE COMMUNICATIONS | (2018) 9:2629 | DOI: 10.1038/s41467-018-05096-6 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05096-6 ARTICLE enzyme (NEB) according to the manufacturer’s instructions. Polyadenosine tails on an Infinite 200Pro Microplate Reader (Tecan) after 45 s. To detect luminescence were added to capped linear transcripts using E. coli PolyA Polymerase (NEB) from Firefly luciferase, 100 µL of Bright-Glo Luciferase reagent (Promega) was according to manufacturer’s instructions, and fully processed mRNA was column added to each well, mixed, and incubated for 5 min. 100 µL of the culture medium purified. For circRNA, after DNase treatment additional GTP was added to a final and luciferase reagent mix was then transferred to a flat-bottomed white-walled concentration of 2 mM, and then reactions were heated at 55 °C for 15 min. RNA plate and luminescence was detected as described above. GFP fluorescence was was then column purified. In some cases, purified RNA was recircularized: RNA detected 24 h post-transfection and images were taken using an EVOS FL cell was heated to 70 °C for 5 min and then immediately placed on ice for 3 min, after imager (Invitrogen). Erythropoietin was detected by solid phase sandwich ELISA which GTP was added to a final concentration of 2 mM along with a buffer (R&D Systems) essentially according to the manufacturer’s instructions except cell including magnesium (50 mM Tris-HCl, 10 mM MgCl2, 1 mM DTT, pH 7.5; New culture supernatant 24 h post transfection was used, and samples were diluted England Biolabs). RNA was then heated to 55 °C for 8 min, and then column 1:200 before use. Interferon-ɑ and interleukin 6 were detected by Fireplex purified. To enrich for circRNA, 20 µg of RNA was diluted in water (86 µL final immunoassay (Abcam). volume) and then heated at 65 °C for 3 min and cooled on ice for 3 min. 20U RNase R and 10 µL of 10× RNase R buffer (Epicenter) was added, and the reaction Flow cytometry. CRISPR-Cas9-mediated GFP ablation was detected by flow was incubated at 37 °C for 15 min; an additional 10U RNase R was added halfway cytometry 96 h after transfection. HEK293-GFP and HEK293 control cells were through the reaction. RNase R-digested RNA was column purified. RNA was trypsinized and suspended in Dulbecco’s Modified Eagle’s Medium (4500 mg/L separated on precast 2% E-gel EX agarose gels (Invitrogen) on the E-gel iBase glucose) supplemented with 10% fetal bovine serum and penicillin/streptomycin. (Invitrogen) using the E-gel EX 1–2% program; ssRNA Ladder (NEB) was used as a Cells were then washed twice in FACS buffer (PBS, 5% heat-inactivated fetal bovine standard. We were unable to obtain adequate circRNA separation using other serum) and resuspended in FACS buffer containing Sytox Blue Dead Cell Stain agarose gel systems. Bands were visualized using blue light transillumination and (Thermo Fisher) according to the manufacturer’s instructions, or FACS buffer quantified using ImageJ. Unprocessed agarose gel images are present as Supple- alone for GFP and blank controls. Fluorescence was detected for 10,000 events on a mentary Fig. 8. Unprocessed agarose gel images including ladders for those gels BD FACSCelesta flow cytometer (BD Biosciences). Data was analyzed in Flowjo from Figs. 1 and 2 that did not include ladders are present as Supplementary Fig. 9. (Flowjo LLC). For gel extractions, bands corresponding to the circRNA were excised from the gel and then extracted using a Zymoclean Gel RNA Extraction Kit (Zymogen). For high-performance liquid chromatography, 30 µg of RNA was heated at 65 °C for 3 Reverse transcription and qPCR. A total of 50,000 HeLa cells were reverse min and then placed on ice for 3 min. RNA was run through a 4.6 × 300 mm size- transfected with 200 ng of GLuc circRNA or modified linear mRNA. Cells were exclusion column with particle size of 5 µm and pore size of 200 Å (Sepax Tech- washed and RNA was harvested and purified 24 h after transfection using an nologies; part number: 215980P-4630) on an Agilent 1100 Series HPLC (Agilent). RNeasy Mini Plus kit (Qiagen) according to the manufacturer’s instructions. RNA was run in RNase-free TE buffer (10 mM Tris, 1 mM EDTA, pH:6) at a flow Synthesis of first-strand cDNA from total RNA was performed with High-Capacity rate of 0.3 mL/minute. RNA was detected by UV absorbance at 260 nm, but was cDNA Reverse Transcription Kit using random hexamers (Thermo Fisher Scien- collected without UV detection. Resulting RNA fractions were precipitated with 5 tific). Gene specific TaqMan primers were purchased as Assay-on-Demand from M ammonium acetate, resuspended in water, and then in some cases treated with (Thermo Fisher Scientific), GAPDH (Hs99999905_m1), DDX58 RNase R as described above. (Hs01061436_m1), IFN-β1 (Hs01077958_s1). The qPCR reaction was carried out using LightCycler 480 Probe Master Mix (Roche) and LightCycler 480 instrument (Roche). For each sample, the real-time PCR reaction was performed in duplicate RNase H nicking analysis. Splicing reactions enriched for circRNA with RNase R and the averages of the obtained threshold cycle values (Ct) were processed for fi and then column puri ed were heated at 65 °C for 5 min in the presence of a DNA further calculations according to the comparative Ct method. Gene expression fi probe (Supplementary Data 1)at ve-fold molar excess, and then annealed at room levels were normalized to the expression of the housekeeping gene GAPDH. temperature. Reactions were treated with RNase H (New England Biolabs) in the provided reaction buffer for 15 min at 37 C. RNA was column purified after digestion. Statistics. Statistical analysis of the results was performed by a two-tailed unpaired Welch’s t-test, assuming unequal variances. Differences were considered significant when p < 0.05. For all studies, data presented is representative of one independent Reverse transcription and cDNA synthesis. For splice junction sequencing, experiment. splicing reactions enriched for circRNA with RNase R and then column purified were heated at 65 °C for 5 min and cooled on ice for 3 min to standardize sec- Data availability fi ondary structure. Reverse transcription reactions were carried out with Superscript . The data that support the ndings of this study are available IV (Invitrogen) as recommended by the manufacturer using a primer specific for a from the corresponding author upon reasonable request. region internal to the putative circRNA. PCR product for sequencing was synthesized using Q5 polymerase (New England Biolabs) and a pair of primers Received: 20 March 2018 Accepted: 13 June 2018 spanning the splice junction.

Tissue culture and transfections. HEK293, HEK293-GFP, HeLa, and A549 cells ’ fi ’ (ATCC) were cultured at 37 °C and 5% CO2 in Dulbecco s Modi ed Eagle s Medium (4500 mg/L glucose) supplemented with 10% heat-inactivated fetal bovine serum (hiFBS, Gibco) and penicillin/streptomycin. HEK293 and HeLa cells tested References negative for mycoplasma. Min6 (a gift from Gordon C. Weir, Joslin Diabetes 1. Barrett, S. P. & Salzman, J. Circular RNAs: analysis, expression and potential Center) medium was additionally supplemented with 5% hiFBS, 20 mM HEPES functions. Development 143, 1838–1847 (2016). μ (Gibco), and 50 M beta-mercaptoethanol (BioRad). Cells were passaged every 2. Chen, L.-L. & Yang, L. Regulation of circRNA biogenesis. RNA Biol. 12, 2–3 days. For all circRNA data sets presented in Fig. 2 except Cas9, 40–100 ng of – fi 381 388 (2015). RNase R-treated splicing reactions or HPLC-puri ed circRNAs were reverse 3. Jeck, W. R. & Sharpless, N. E. Detecting and characterizing circular RNAs. transfected into 10,000 HEK293 cells/100 µL per well of a 96-well plate using Nat. Biotechnol. – ’ 32, 453 461 (2014). Lipofectamine MessengerMax (Invitrogen) according to the manufacturer s 4. Wang, Y. & Wang, Z. Efficient backsplicing produces translatable circular instructions. For Cas9, 100 ng of in vitro transcribed sgRNA was reverse trans- mRNAs. RNA 21, 172–179 (2015). fected alone or cotransfected with 150 ng of RNase R-treated Cas9 splicing reaction 5. Hansen, T. B. et al. Natural RNA circles function as efficient microRNA into 50,000 HEK293-GFP cells/500uL per well of a 24-well plate using Messen- sponges. Nature 495, 384–388 (2013). gerMax. For all RNA datasets presented in Fig. 3, equimolar quantities of each 6. Li, Z. et al. Exon-intron circular RNAs regulate transcription in the nucleus. RNA (equivalent to between 40 and 80 ng dependent on size) were reverse Nat. Struct. Mol. Biol. – transfected into 10,000 HEK293, HeLa, or A549 cells/100 uL per well of a 96-well 22, 256 264 (2015). plate using MessengerMax. For experiments wherein protein expression was 7. Legnini, I. et al. Circ-ZNF609 is a circular RNA that can be translated and functions in myogenesis. Mol. Cell 66,22–37.e9 (2017). assessed at multiple time points, media was fully removed and replaced at each Mol. Cell – time point. Min6 cells were transfected in 96-well plate format between 60–80% 8. Pamudurti, N. R. et al. Translation of CircRNAs. 66,9 21.e7 (2017). confluency. Sample sizes were chosen based on pilot experiments to determine 9. Enuka, Y. et al. Circular RNAs are long-lived and display only minimal early Nucleic Acids Res. – assay variance and to minimize reagent consumption while allowing for mean- alterations in response to a growth factor. 44, 1370 1383 ingful differences between conditions to be distinguished. (2016). 10. Kaczmarek, J. C., Kowalski, P. S. & Anderson, D. G. Advances in the delivery of RNA therapeutics: from concept to clinical reality. Genome Med. 9,60 Protein expression analysis. For luminescence assays, cells and media were (2017). harvested 24 h post-transfection. To detect luminescence from Gaussia luciferase, 11. Fink, M., Flekna, G., Ludwig, A., Heimbucher, T. & Czerny, T. Improved – fl 10 20 µL of tissue culture medium was transferred to a at-bottomed white-walled translation efficiency of injected mRNA during early embryonic development. plate (Corning). 25 µL of BioLux Gaussia Luciferase reagent including stabilizer Dev. Dyn. 235, 3370–3378 (2006). (New England Biolabs) was added to each sample and luminescence was measured

NATURE COMMUNICATIONS | (2018) 9:2629 | DOI: 10.1038/s41467-018-05096-6 | www.nature.com/naturecommunications 9 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05096-6

12. Ferizi, M. et al. Stability analysis of chemically modified mRNA using Acknowledgements micropattern-based single-cell arrays. Lab Chip 15, 3561–3571 (2015). This work was supported in part by the Defense Advanced Research Projects Agency 13. Holtkamp, S. et al. Modification of antigen-encoding RNA increases stability, (Grant W32P4Q-13-1-0011). P.K. acknowledges funding from the Juvenile Diabetes translational efficacy, and T-cell stimulatory capacity of dendritic cells. Blood Research Foundation (JDRF) postdoctoral fellowship (Grant 3-PDF-2017-383-A-N). We 108, 4009–4017 (2006). also thank the Nanotechnology Materials and Flow Cytometry Core Facilities at the MIT 14. Kuhn, A. N. et al. Phosphorothioate cap analogs increase stability and Koch Institute. translational efficiency of RNA vaccines in immature dendritic cells and Gene Ther. – induce superior immune responses in vivo. 17, 961 971 (2010). Author contributions 15. Presnyak, V. et al. Codon optimality is a major determinant of mRNA stability. Cell 160, 1111–1124 (2015). R.W. and P.K. designed and executed experiments and wrote the manuscript. D.A. wrote 16. Kariko, K., Muramatsu, H., Ludwig, J. & Weissman, D. Generating the the manuscript and provided guidance. optimal mRNA for therapy: HPLC purification eliminates immune activation and improves translation of nucleoside-modified, protein-encoding mRNA. Additional information Nucleic Acids Res. 39, e142–e142 (2011). Supplementary Information accompanies this paper at https://doi.org/10.1038/s41467- 17. Petkovic, S. & Müller, S. RNA circularization strategies in vivo and in vitro. 018-05096-6. Nucleic Acids Res. 43, 2454–2465 (2015). 18. Beadudry, D. & Perreault, J.-P. An efficient strategy for the synthesis of Competing interests: D.A., P.K., and R.W. filed a patent for the development of the circular RNA molecules. Nucleic Acids Res. 23, 3064–3066 (1995). circRNA technology. 19. Micura, R. Cyclic oligoribonucleotides (RNA) by solid-phase synthesis. Chem. A Eur. J. 5, 2077–2082 (1999). Reprints and permission information is available online at http://npg.nature.com/ 20. Puttaraju, M. & Been, M. Group I permuted intron-exon (PIE) sequences self- reprintsandpermissions/ splice to produce circular exons. Nucleic Acids Res. 20, 5357–5364 (1992). 21. Ford, E. & Ares, M. Synthesis of circular RNA in bacteria and yeast using Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in RNA cyclase ribozymes derived from a group I intron of phage T4. Proc. Natl published maps and institutional affiliations. Acad. Sci. 91, 3117–3121 (1994). 22. Vicens, Q., Paukstelis, P. J., Westhof, E., Lambowitz, A. M. & Cech, T. R. Toward predicting self-splicing and protein-facilitated splicing of group I introns. RNA 14, 2013–2029 (2008). Open Access This article is licensed under a Creative Commons 23. Borman, A. M., Le Mercier, P., Girard, M. & Kean, K. M. Comparison of Attribution 4.0 International License, which permits use, sharing, picornaviral IRES-driven internal initiation of translation in cultured cells of adaptation, distribution and reproduction in any medium or format, as long as you give Nucleic Acids Res. – different origins. 25, 925 932 (1997). appropriate credit to the original author(s) and the source, provide a link to the Creative fi 24. Imataka, H., Gradi, A. & Sonenberg, N. A newly identi ed N-terminal amino Commons license, and indicate if changes were made. The images or other third party acid sequence of human eIF4G binds poly(A)-binding protein and functions material in this article are included in the article’s Creative Commons license, unless EMBO J. – in poly(A)-dependent translation. 17, 7480 7489 (1998). indicated otherwise in a credit line to the material. If material is not included in the 25. Kahvejian, A. Mammalian poly(A)-binding protein is a eukaryotic translation ’ Genes Dev. – article s Creative Commons license and your intended use is not permitted by statutory initiation factor, which acts via multiple mechanisms. 19, 104 113 regulation or exceeds the permitted use, you will need to obtain permission directly from (2005). the copyright holder. To view a copy of this license, visit http://creativecommons.org/ 26. Weingarten-Gabbay, S. et al. Comparative genetics. Systematic discovery of licenses/by/4.0/. cap-independent translation sequences in human and viral genomes. Science 351, aad4939 (2016). 27. Li, Y. & Breaker, R. R. Kinetics of RNA degradation by specific base catalysis © The Author(s) 2018 of transesterification involving the 2‘-hydroxyl group. J. Am. Chem. Soc. 121, 5364–5372 (1999). 28. Chen, Y. G. et al. Sensing self and foreign circular RNAs by intron identity. Mol. Cell 67, 228–238.e5 (2017).

10 NATURE COMMUNICATIONS | (2018) 9:2629 | DOI: 10.1038/s41467-018-05096-6 | www.nature.com/naturecommunications